Author: Uri Blass
Date: 02:17:40 12/22/03
Go up one level in this thread
On December 22, 2003 at 00:06:35, Robert Hyatt wrote: >On December 21, 2003 at 23:04:12, Mike Byrne wrote: > >>On December 21, 2003 at 22:53:25, Sean Mintz wrote: >> >>>I am curious how many people run Crafty with /DDETECTDRAW set; and more >>>specifically, if anyone has run any matches to see how it affects crafty's >>>strength. >>> >>>Sean Mintz >> >>The versions I have made recently had it set. > > >Someone sent me a position where it screwed up pretty badly. I have not had >time to look at it carefully however. If you are interested (I don't think >he will mind me posting this email here): > > >The second problem : >I noticed a position where the DETECTDRAW code seems to fail badly. >In the following sample, Crafty (compiled with DETECTDRAW) detects draw = >when there is not any draw at all. >In this position Crafty sacrify a piece (playing Nd3) to detect = >erroneously a draw and lose the game badly. >You will see the logs files below for the two cases. >=20 > >CRAFTY WITH DETECTDRAW plays Nd3 : > >White(1): setboard 3r1rk1/p1p4p/1p2b1pP/2n1PpP1/q1P2B2/4QP2/P6R/2R2BK1 b = >- - 0 25 I am not surprised. I already said that it is more logical to use detect draw to reduce the score of the stronger side and not to have draw score because it is obvious that it does not always detect correct draws. I do not understand why people choose extreme solutions. The score should be a function of the expected result based on the knowledge of the program. The fact that crafty detects some suspect to draw that it is not clear if the suspect is correct should make the expected result closer to 0.50 but not 0.50 Uri
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.